Map fusion method suitable for sub-map with few overlapping parts

Abstract

The invention relates to the technical field of map fusion, in particular to a map fusion method suitable for a sub-map with few overlapping parts. The method comprises: firstly, iteratively calculating a rotation matrix R by using point cloud plane extraction, normal vector estimation and normal vector projection angle statistics methods, then iteratively transforming source point clouds by usingthe R, finally, performing three-view projection on the two groups of point clouds to obtain two groups of grey-scale maps, and calculating a translation vector t by using a phase correlation methodfor the grey-scale maps; and finally obtaining a transformation matrix (R, t) between the two groups of point clouds. According to the map fusion method suitable for a sub-map with few overlapping parts, the fusion of the sub-maps under the low overlapping degree can be correctly completed, the error of searching corresponding points by the previous methods is avoided, and the method is suitable for the registration between two groups of large map point clouds with fewer overlapping parts and three axes having rotating angles at the same time.

Description

technical field [0001] The present invention relates to the field of map fusion technology, and more particularly, to a map fusion method suitable for less overlapping parts of sub-maps. Background technique [0002] In the tasks of performing surveying and mapping and exploring unknown environments, it is usually possible to send a single or multiple unmanned vehicles to build maps to improve efficiency; but when there is a lack of location information such as GPS, how to collect single or multiple unmanned vehicles The fusion of the obtained point cloud map data has become a difficult problem. At this time, point cloud map fusion technology is particularly important. [0003] The most commonly used point cloud registration algorithm is the iterative nearest neighbor algorithm represented by ICP. The main step in this type of method is to find the closest point pair in the two sets of point clouds, and according to the estimated transformation relationship (R and t) To ca...

AI Technical Summary

Problems solved by technology

The two types of point-based geometric feature registration methods also have their own shortcomings: the four-point method is better for two sets of point clouds with less overlapping parts, and the registration effect is better when the amount of point cloud data is small, but for large-scale map point clouds. The registration effect is poor. The reason is that the four-point method determines the corresponding points by finding the affine invariance of two sets of point cloud points that are not coplanar. Small point clouds can correctly determine the corresponding points. For large map point clouds, the determined corresponding points It is not necessarily an overlapping part, so the registration effect is poor

This method can register two sets of point clouds with less overlapping parts to a certain extent, but it still has defects. It is only suitable for the case where there is only z-axis rotation between the two sets of point clouds.

The point cloud registration scheme based on statistical histograms can register two sets of point clouds with less overlapping parts to a certain extent, but it is only suitable for the case where there is only z-axis rotation between the two sets of point clouds.

There are two main reasons. One is that this method uses all points in the map to calculate and project the normal vector. When calculating the arctangent of the normal vector projection, a large number of zero and infinite values ​​will affect the angle statistics; the other is that in the three-axis When there are rotation angles at the same time, the algorithm calculates Euler angles in the order of z-y-x axes, which will cause the error generated by the previous angle calculation to affect the subsequent angle calculation, and because the iteration step is set, the error increases with the number of iterations. , so it cannot be correctly registered

Images